Vladimir MATIĆ, Aleksandar LEBL, Dragan MITIĆ, Miroslav DUKIĆ

Institute for Telecommunications and Electronics IRITEL A.D. BELGRADE, Faculty of Electrical Engineering Belgrade

Estimation of Post Dialling Delay in Telephone Networks

Abstract. In this paper we present that Post Dialling Delay in one telephone network depends on the service discipline in particular network nodes. It
is indicated that deviation of PDD from its mean value decreases as the number of nodes included in one established connection increases. The
consequence of this fact is that the condition of serving 95% of connections is more stringent in the connections with small number of network nodes
included in the connection, while the condition dealing with the mean value of PDD is more stringent in the connections with great number of nodes.

Streszczenie. W artykule analizowane jest opóźnienie połączenia w sieciach telefonicznych w zależności od liczby węzłów i liczby połączeń.
(Określenie opóźnienia połączenia w sieciach telefonicznych)

Keywords: Telephone signalling, Post dialling delay, telephone networks

Słowa kluczowe: sieci telefoniczne, opóźnienie połaczenia.

Introduction values are determined for ISDN network, but they are also
The speed of telephone network reaction is, besides applied in mixed, [2], and packet networks [3]. In this paper
voice quality, the main indicator of Quality of Service (QoS). we try to indicate the possibility to estimate PDD in
The speed of network reaction is expressed by the time depending on the type of service in network nodes.
while the activities, such as connection realization to the
called side, transferring the signal of called subscriber Definition and recommendations
answer, connection disconnect, etc., are accomplished. For all types of networks, PDD can be defined as the
The speed of these activities is measured by the time time interval from the end of sending user signalling
interval needed for their realization, and this time intervals address information till the beginning of the called side
are called Post Dialing Delay, Answer Signal Delay, Call response, Fig. 1. This definition is detailed for some
Release Delay, etc. Among these indicators of function networks, where PDD is defined as the time interval
speed, it seems that the most important one is Post Dialing between specific signalling messages (from the message
Delay (PDD), which is also called Post Selection Delay. It is SETUP till the message ALERTING in ISDN) or methods
the key indicator of speed of the connection realization for (from the method INVITE till the answer 180 RINGING in
two reasons. This phase is technically the most SIP signalling, [4]). We adopt that PDD in SIP signalling
complicated one (1), and in this phase the user expects the ends by the preliminary message 180 RINGING, rather than
positive answer, i.e. he is prone to abandon from the the preliminary message 100 TRYING, as stated in [4], or
attempt to establish the connection (2). The greatest values the final message 200 OK, as stated in [5].
of this time are suggested in Recommendation [1]. These

Fig.1. Presentation of the PDD time interval

In order to achieve the good service for the user, the passes through 1-4, 5-7 and 8-10 network nodes,
recommendations prescribe the values of PDD. These respectively. It can be noticed that the ratio of the greatest
recommendations determine: the longest mean time of PDD recommended values t95PDD and tmPDD is t95PDD/tmPDD=2 (for
(tmPDD) and the longest PDD duration for which 95% of local connections), t95PDD/tmPDD=1.6 (for transit connections),
connections will receive the answer from the called side t95PDD/tmPDD=1.375 (for international connections).
(t95PDD).
The recommendations are defined for local, transit and What elements constitute PDD
international connections. In [1] and [2] it is stated that Interval PDD can be presented as the sum of 2·s time
tmPDD≤3s, 5s and 8s, and t95PDD≤6s, 8s and 11s for local, intervals, concerning one network node and information
transit and international connections (respectively) and transfer to the next network node. These time intervals exist
normal traffic load (load A). For the increased traffic load for the transfer from the calling to the called user, and vice
(load B) the values tmPDD≤4.5s, 7.5s and 12s, and versa, Fig. 1. Actions performed in one node and on one
t95PDD≤9s, 12s and 16.5s for local, transit and international link (i.e. on one section) in literature are called subcall or
connections (respectively) are recommended. In [1] it is call segment, [6]. The time needed for execution of one call
stated that typical local, transit and international connection segment is random variable, which depends on the type of

154 PRZEGLĄD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 5b/2012

operation in network node and on link. In some nodes it is
necessary to access the base. It can be said that the
probability distribution of one call segment duration time
depends on the service discipline in the node and on the
number of service elements in section. In the next section
we shall see what main distributions could exist.
Servicing in one section
1. Servicing by one server
Let us suppose that the time interval, which address
signalling message spends in network node and on the link
towards the next network node, is caused by only one
bottleneck, i.e. by waiting time on processor service of
Fig. 2. Cumulative distribution function of service time for M/D/1
signalling message or by the waiting on sending over link. and M/M/1 system when A=0.9
The time interval of signalling message processing (service
time, ts) is either constant, or is negative exponentially If link is the main reason of waiting, then the service
distributed. (Precisely speaking, the time interval spent in time is proportional to the duration of the packets, which are
the node is the sum of service time and the waiting time. In sent over the link. If the signalling processor is the element,
the periods of heavy load, which are especially interesting which causes waiting, then the dispersion of service time for
for us, service time can be neglected in relation to waiting particular signalling messages can be greater than the
time). We shall consider service according to FIFO dispersion of the duration of particular messages.
discipline. In the first case (constant service time), the
queueing system is M/D/1, and in the second case 2. Multiphase servicing in one section
(negative exponentially distributed service time), queueing Let us consider one network node and one link, i.e. one
system is M/M/1. It means that all serving resources except section in the case that congestion can happen on several
one are over-dimensioned, i.e. they provide service without elements: signalling processor, data base, sending buffer.
significant waiting. The waiting is caused by offered traffic A In this case, the time, which signalling information spends in
on only one server and can be presented by random the considered node and link is the sum of service time and
variable TD or TM. As it is well known and explained in [7], waiting time in each element of service in that node (link). In
the mean waiting time, tmD, and the probability of waiting general circumstances, it is very complicate to calculate this
time. That’s why two assumptions are adopted. The first
longer than t, P(TD>t), is in the case of M/D/1 system: one is that the service of the same kind is performed in
ts  A each service system of one section, and the second one is
(1) tmD  )
2  (1  A) that the time spent on one section (node + link) is the
random variable, which is the sum of k random variables
t representing the time spent for particular phases in the
l ( A  (  i ))i A( t i )
ts node and on the link, where k is the number of service
(2) P (TD  t )  1  (1  A)   i!
 e ts phases on one section (k=2, 3, 4…). Let us suppose that in
i 0 each service phase of one section the time spent in that
where l is the integer part of the ratio t/ts. phase is random variable with exponential distribution of the
In the case of M/M/1 system, the mean waiting time tmM time duration and mean value 1/λ. In each phase a lot of
and the probability of waiting P(TM>t) are: different, independent of each other, requests are serviced.
That’s why the distribution of total time TE, spent on one
ts  A section with k phases, is well known Erlang-k distribution,
(3) tmM 
(1  A) expressed by the equation
k 1
t (  t ) i
(4) P (TM  t )  A  e
 (1 A)
ts (5) PE (TE  t )   e t  i!
i 0
The characteristics of serving by one server are: The mean value tmE of random variable TE for which
- probability distribution function of waiting time is holds Erlang-k distribution is k/λ.
exponential; For this distribution, there exists one value t95E that
- mean waiting time is shorter in the case of M/D/1 holds
model than in the case of M/M/1 model, but model M/M/1 1  PE (TE  t95E )  PE (TE  t95E )  0, 95
appears more often in practical systems;
- the ratio t95/tm for both models is about 3, and,
according to [7], the more stringent criterion, which is
recommended in [8] and [9], is the one dealing with t95.
Fig. 2. presents cumulative distribution function in these
two systems for traffic load A=0.9. On the x axis the service
time ts is used as the unit of time.
Therefore, if the delay in one network node is the
consequence of the service in only one element (processor,
link), then exponential distribution is valid for cumulative
distribution function. The ratio of t95 to tm is about 3, but the
absolute values for constant service time are significantly
lower. Unfortunately, constant service time is rare in the
practice. We can note that, in this case, is more difficult to
satisfy the criterion t95 than the criterion tm, [8]. Fig. 3. Ratio t95E/tmE in the function of number of service phases on
one section

PRZEGLĄD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 5b/2012 155

 Fig. 3. presents the ratio t95E/tmE for the total time spent answer from the called user side. The answers from the
on one section as the function of the number of service particular points of the network can’t be considered.
phases. PDD is defined by its mean value tmPDD and the time
Comparing the cases 1. (one phase) and 2. (multiphase) t95PDD while the answer is received from the called side in
when the mean time spent on one section is same, it can be 95% occurrences. The recommendations dealing with PDD
said that the service in few phases is more favourable, also consider these two time intervals. It can be concluded
because the dispersion, i.e. deviation of the time spent on that the mean value tmPDD increases proportionally to the
one section from its mean value is less than in case of number of sections in the network, which signalling
service on one channel. message passes. Comparing to the increasing of mean
value of PDD, the deviation of PDD from its mean value
Total duration of time interval PDD increases slower as the number of sections increases.
Time interval PDD is, obviously, random variable, that is This increasing of PDD deviation is greatest for local
the sum of mutually independent random values of call connections (2 to 8 sections), and smallest for international
segment durations. It must be indicated that PDD is connections (16 to 20 sections).
constituted of the call segments of the messages sent The increasing of PDD deviation from its mean value is
forward (e.g. SETUP, IAM (Initial Address Message), additionally reduced if in network sections multistep
INVITE) and backward (e.g. ALERT, ACM (Address (multiphase) service is performed.
Complete Message), 180 RINGING), i.e. 2·s segments, Fig. In the connections with small number of sections it is
1. (However, PDD is mainly constituted of sending more difficult to satisfy the recommendations dealing with
messages forward, because these messages are more the greatest allowed value of t95PDD. In opposite, in the case
complicate and require the decision about forwarding and of great number of sections, the recommendations dealing
acceptance in each node). Therefore, the characteristics of with the greatest value of tmPDD become more stringent.
PDD will be the same as the characteristics of random
variable, which is the sum of 2·s independent random REFERENCES
variables. As is well known, the mean time PDD will be [1] ITU-T: Recommendation E.721 - Network grade of service
equal to the sum of all mean durations of particular call parameters and target values for circuit-switched services in
segments, i.e. the evolving ISDN, May 1999
[2] ITU-T: Recommendation E.671 - Post-selection delay in
2 s
 tmi
PSTN/ISDN networks using Internet telephony for a portion of
(6) tmPDD  the connection, March 2000
i 1 [3] Arjona, A., Westphal, C., Yla-Jaaski, A., Kristensson, M. and
Manner, J.: Towards High Quality VoIP in 3G Networks an
What will be the duration t95 of PDD, i.e. t95PDD? In general Empirical Approach, Int. J. of Communications, Network and
situation the calculation of this value will be very complicate. System Sciences, 1(2008), No 4, pp 350-361
But, as we know that [4] Eyers, T. and Schulzrinne, H.: Predicting Internet Telephony
2 s Call Setup Delay, Proc. IP Telephony Workshop, April 2000
(7)  2 PDD    i 2 [5] Pack, S. and Lee, H.: Call Setup Latency Analysis in SIP-Based
Voice over WLANs, IEEE Communications Letters, 12(2008),
i 1
No 2, pp. 103-105
[6] Lin, H., Seth, T., Broscius, A. and Huitema, C.: VoIP Signaling
and starting from the fact that is for each distribution the Performance Requirements and Expectations, Internet draft,
time interval t95 proportional (~) to the standard deviation σ IETF, October 1999
(t95 ~ σ), we can conclude that t95PDD is proportional to the [7] Markov, Ž. and Manević, I.: Determination of More Stringent
square root of the number of sections, i.e. Criterion for Common Control Unit of Digital Telephone
Exchange, Int. J. Electron. Commun. (AEU) 52(1998), No 2, pp
(8) t95PDD  2  s 101-103
[8] ITU-T: Recommendation Q.543 - Digital exchange design
objectives, March 1993
When the number of sections increases, the ratio [9] ITU-T: Recommendation Q.725 - Signaling System No7 –
t95PDD/tmPDD decreases, because it is signaling performance in the telephone application, March
1993
t95PDD 2s 1
(9)   ,s  1
tmPDD 2s 2s Authors: mr Vladimir Matić, dipl. Ing., IRITEL A.D., Batajnički put
23, 11080 Belgrade, Serbia, (phone 381-11-3073485; e-mail:
It can be concluded that the deviation of time interval [email protected]; dr Aleksandar Lebl dipl.ing., IRITEL A.D.,
PDD from its mean value is smaller if the number of Batajnički put 23, 11080 Belgrade, Serbia, (phone 381-11-
segments is greater. If in each network section the 3073422; e-mail: [email protected]; dr Dragan Mitić dipl.ing., IRITEL
processing of signalling information is subject of multistep A.D., Batajnički put 23, 11080 Belgrade, Serbia, phone 381-11-
service (subsection 2.), this effect is even more expressed. 3073420; e-mail: [email protected]; prof. dr Miroslav Dukić dipl.ing.,
Faculty of Electrical Engineering, Bulevar Kralja Aleksandra 73,
11000 Belgrade, Serbia, phone 381-11-3370106; e-mail:
Conclusions [email protected].
PDD is the time interval calculated from sending
address information towards called user till receiving the

156 PRZEGLĄD ELEKTROTECHNICZNY (Electrical Review), ISSN 0033-2097, R. 88 NR 5b/2012